[Xen-devel] Xen NUMA memory allocation policy

[Saurabh Mishra <saurabh.globe@xxxxxxxxx>]

Hi --

We are using Xen 4.2.2_06 on SLES SP3 Updates and wanted to know if there is a simple way to gather information about physical pages allocated for a HVM guest. We are trying to figure whether XL is better off in allocating contiguous huge/large pages for a guest or XM. I guess it does not matter since Xen's hypervisor would be implementing page allocation polices.

With xl debug-key u, we know how much memory was allocated from each NUMA node, but we would also like to know whether how much of them were huge pages and were they contiguous or not. Basically we need to retrieve machine pfn and VM's pfn to do some comparison.

(XEN) Memory location of each domain:

(XEN) Domain 0 (total: 603765):

(XEN)     Node 0: 363652

(XEN)     Node 1: 240113

(XEN) Domain 1 (total: 2096119):

(XEN)     Node 0: 1047804

(XEN)     Node 1: 1048315

(XEN) Domain 2 (total: 25164798):

(XEN)     Node 0: 12582143

(XEN)     Node 1: 12582655

We would like Xen to allocate as many as continuous huge/large pages for a HVM guest. So if there is a tunable, then please let me know. The HW is x86-64.

Looks like 'xl debug-key D' can tell us gfn and mfn. What does 'order' and 'is_pod' mean in the output? How can we ensure pages are contiguous?

(XEN) gfn: 180ee00          mfn: 1103800           order:  9  is_pod: 0

(XEN) gfn: 180f000           mfn: 8ad400            order:  9  is_pod: 0

(XEN) gfn: 180f200           mfn: 1103600           order:  9  is_pod: 0

(XEN) gfn: 180f400           mfn: 8ad200            order:  9  is_pod: 0

(XEN) gfn: 180f600           mfn: 1103400           order:  9  is_pod: 0

Thanks,

/Saurabh

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel